Machine-learning-based ground sink susceptibility evaluation using underground pipeline data in Korean urban area.

Ground subsidence caused by natural factors, including groundwater, has been extensively researched. However, there have been few studies on ground sink caused mainly by artifacts, including underground pipelines in urban areas. This paper proposes a method of predicting ground sink susceptibility caused by underground pipelines. Underground pipeline data, drilling data, and 77 points of ground sink occurrence were collected for five 1 × 1 km urban areas. Furthermore, three ground sink conditioning factors (GSCFs) (pipe deterioration, diameter, and length) were identified by correlation analysis. Pipe deterioration showed the highest correlation with ground sink occurrence, followed by pipe length and pipe diameter in that order. Next, four machine learning methods [multinomial logistic regression (MLR), decision tree (DT) classifier, random forest (RF) classifier, and gradient boosting (GB) classifier] were applied. The results show that GB classifier had the highest accuracy of 0.7432, whereas the accuracy of RF classifier was 0.7407; thus, GB classifier was not significantly more accurate. RF classifier showed the highest reliability (0.84, 0.70, 0.87) according to the area under the receiver operating characteristic (AUC-ROC) curve. Ground sink susceptibility maps (GSSMs) of the five regions in an urban area were created using RF classifier, which performed the best overall.

3D convolutional neural network for machining feature recognition with gradient-based visual explanations from 3D CAD models.

In the manufacturing industry, all things related to a product manufactured are generated and managed with a three-dimensional (3D) computer-aided design (CAD) system. CAD models created in a 3D CAD system are represented as geometric and topological information for exchange between different CAD systems. Although 3D CAD models are easy to use for product design, it is not suitable for direct use in manufacturing since information on machining features is absent. This study proposes a novel deep learning model to recognize machining features from a 3D CAD model and detect feature areas using gradient-weighted class activation mapping (Grad-CAM). To train the deep learning networks, we construct a dataset consisting of single and multi-feature. Our networks comprised of 12 layers classified the machining features with high accuracy of 98.81% on generated datasets. In addition, we estimated the area of the machining feature by applying Grad-CAM to the trained model. The deep learning model for machining feature recognition can be utilized in various fields such as 3D model simplification, computer-aided engineering, mechanical part retrieval, and assembly component identification.

Machining feature recognition based on deep neural networks to support tight integration with 3D CAD systems.

Recently, studies applying deep learning technology to recognize the machining feature of three-dimensional (3D) computer-aided design (CAD) models are increasing. Since the direct utilization of boundary representation (B-rep) models as input data for neural networks in terms of data structure is difficult, B-rep models are generally converted into a voxel, mesh, or point cloud model and used as inputs for neural networks for the application of 3D models to deep learning. However, the model's resolution decreases during the format conversion of 3D models, causing the loss of some features or difficulties in identifying areas of the converted model corresponding to a specific face of the B-rep model. To solve these problems, this study proposes a method enabling tight integration of a 3D CAD system with a deep neural network using feature descriptors as inputs to neural networks for recognizing machining features. Feature descriptor denotes an explicit representation of the main property items of a face. We constructed 2236 data to train and evaluate the deep neural network. Of these, 1430 were used for training the deep neural network, and 358 were used for validation. And 448 were used to evaluate the performance of the trained deep neural network. In addition, we conducted an experiment to recognize a total of 17 types (16 types of machining features and a non-feature) from the B-rep model, and the types for all 75 test cases were successfully recognized.